SHAH NAWAZ

In my research on a privacy-preserving federated learning framework for lung CT analysis, I designed a multi-client architecture for medical collaboration without data sharing. I implemented U-Net segmentation that achieved a 0.8568 Dice score and evaluated robustness under non-IID client data distributions.

For JaaLTaKa, I developed an explainable system for counterfeit Bangladeshi banknote detection, using ResNet50 feature extraction and Transformer-based view fusion. It achieved 99.04% accuracy, and I deployed the model to mobile with ONNX Runtime and INT8 quantization.

I also developed an LLM-based codebase assistant with embedding-based indexing, local and API-based model inference, and a REST backend. I'm seeking an MSc in Data Science and Artificial Intelligence to strengthen my foundations in statistical learning and distributed AI systems, with long-term research goals in trustworthy and interpretable machine learning.